def train():
"""Train."""
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
dataset = DataSet(FLAGS.batch_size)
images, depths, invalid_depths = dataset.csv_inputs(FLAGS.train_file)
keep_conv = tf.placeholder(tf.float32)
keep_hidden = tf.placeholder(tf.float32)
if FLAGS.refine_train:
print("refine train.")
coarse = model.inference(images, keep_conv, trainable=False)
logits = model.inference_refine(images, coarse, keep_conv,
keep_hidden)
else:
print("coarse train.")
logits = model.inference(images, keep_conv, keep_hidden)
loss = model.loss(logits, depths, invalid_depths)
train_op = op.train(loss, global_step, FLAGS.batch_size)
init_op = tf.initialize_all_variables()
# Session
# sess = tf.Session(config=tf.ConfigProto(
# log_device_placement=FLAGS.log_device_placement,
# device_count={'GPU': 1}))
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement,
gpu_options=tf.GPUOptions(visible_device_list='1')))
merged = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(FLAGS.log_dir, sess.graph)
sess.run(init_op)
# parameters
coarse_params = {}
refine_params = {}
if FLAGS.refine_train:
for variable in tf.all_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
print("parameter: %s" % (variable_name))
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
else:
for variable in tf.trainable_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
# define saver
print coarse_params
saver_coarse = tf.train.Saver(coarse_params)
if FLAGS.refine_train:
saver_refine = tf.train.Saver(refine_params)
# fine tune
if FLAGS.fine_tune:
coarse_ckpt = tf.train.get_checkpoint_state(FLAGS.coarse_dir)
if coarse_ckpt and coarse_ckpt.model_checkpoint_path:
print("Pretrained coarse Model Loading.")
saver_coarse.restore(sess, coarse_ckpt.model_checkpoint_path)
print("Pretrained coarse Model Restored.")
else:
print("No Pretrained coarse Model.")
if FLAGS.refine_train:
refine_ckpt = tf.train.get_checkpoint_state(FLAGS.refine_dir)
if refine_ckpt and refine_ckpt.model_checkpoint_path:
print("Pretrained refine Model Loading.")
saver_refine.restore(sess,
refine_ckpt.model_checkpoint_path)
print("Pretrained refine Model Restored.")
else:
print("No Pretrained refine Model.")
# train
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
for step in xrange(FLAGS.max_steps):
index = 0
for i in xrange(1000):
summary, _, loss_value, logits_val, images_val = sess.run(
[merged, train_op, loss, logits, images],
feed_dict={
keep_conv: 0.8,
keep_hidden: 0.5
})
if index % 10 == 0:
print("%s: %d[epoch]: %d[iteration]: train loss %f" %
(datetime.now(), step, index, loss_value))
summary_writer.add_summary(summary, 1000 * step + i)
assert not np.isnan(
loss_value), 'Model diverged with loss = NaN'
if index % 500 == 0:
if FLAGS.refine_train:
output_predict(
logits_val, images_val,
os.path.join(
FLAGS.output_dir,
"predict_refine_%05d_%05d" % (step, i)))
else:
output_predict(
logits_val, images_val,
os.path.join(FLAGS.output_dir,
"predict_%05d_%05d" % (step, i)))
index += 1
if step % 5 == 0 or (step * 1) == FLAGS.max_steps:
if FLAGS.refine_train:
refine_checkpoint_path = FLAGS.refine_dir + '/model.ckpt'
saver_refine.save(sess,
refine_checkpoint_path,
global_step=step)
else:
coarse_checkpoint_path = FLAGS.coarse_dir + '/model.ckpt'
saver_coarse.save(sess,
coarse_checkpoint_path,
global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()
def train():
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
dataset = DataSet(BATCH_SIZE)
# Get batch of tensors.
# images is X, depth is Y, invalid_depth is matrix of 0 and 1
# that indicates which pixel in Y has depth value.
images, depths, invalid_depths = dataset.csv_inputs(TRAIN_FILE)
# keep_conv is the drop out rate in conv layers
keep_conv = tf.placeholder(tf.float32)
# TODO(xuguo): what's keep hidden?
keep_hidden = tf.placeholder(tf.float32)
if REFINE_TRAIN:
# When training with refined network, train with both coarse and
# refined together.
print("refine train.")
coarse = model.inference(images, keep_conv, trainable=True)
logits = model.inference_refine(images, coarse, keep_conv,
keep_hidden)
else:
# When training with coarse network, train with only coarse network.
# (mpng) this isn't called at all
print("coarse train.")
logits = model.inference(images, keep_conv, keep_hidden)
# define loss function:
# logits: the final output after FC layer.
# depth: the Y-hat.
# invalid_depth: the pixels without depth value.
loss = model.loss(logits, depths, invalid_depths)
# define trainning function, adam optimization.
train_op = op.train(loss, global_step, BATCH_SIZE)
# initialize all variable
init_op = tf.global_variables_initializer(
) #tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=LOG_DEVICE_PLACEMENT))
sess.run(init_op)
# parameters
# the parameters are use to store checkpoint, so training can resume when exception
# happens.
coarse_params = {}
refine_params = {}
print('pre-trained parameters -----------------------------------')
if REFINE_TRAIN:
for variable in tf.global_variables(): #tf.all_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
print("parameter: %s" % (variable_name))
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
else:
for variable in tf.trainable_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
# define saver
# print(coarse_params)
saver_coarse = tf.train.Saver(coarse_params)
if REFINE_TRAIN:
saver_refine = tf.train.Saver(refine_params)
# fine tune
if FINE_TUNE:
coarse_ckpt = tf.train.get_checkpoint_state(COARSE_DIR)
if coarse_ckpt and coarse_ckpt.model_checkpoint_path:
print("Pretrained coarse Model Loading.")
saver_coarse.restore(sess, coarse_ckpt.model_checkpoint_path)
print("Pretrained coarse Model Restored.")
else:
print("No Pretrained coarse Model.")
if REFINE_TRAIN:
refine_ckpt = tf.train.get_checkpoint_state(REFINE_DIR)
if refine_ckpt and refine_ckpt.model_checkpoint_path:
print("Pretrained refine Model Loading.")
saver_refine.restore(sess,
refine_ckpt.model_checkpoint_path)
print("Pretrained refine Model Restored.")
else:
print("No Pretrained refine Model.")
# train with multi-thread.
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
loss_curve = []
# step is epoch.
for step in range(EPOCH):
for i in range(TOTAL_BATCH):
_, loss_value, logits_val, images_val = sess.run(
[train_op, loss, logits, images],
feed_dict={
keep_conv: 0.8,
keep_hidden: 0.5
})
if i % 10 == 0:
loss_curve.append(loss_value)
print("%s: %d[epoch]: %d[iteration]: train loss %f" %
(datetime.now(), step, i, loss_value))
assert not np.isnan(
loss_value), 'Model diverged with loss = NaN'
# save parameters every 10 epoch.
if step % 10 == 0 or (step * 1) == EPOCH:
with open('loss.txt', 'w') as output:
for l in loss_curve:
output.write(str(l) + '\n')
if REFINE_TRAIN:
output_predict(logits_val, images_val,
"data/perdict/predict_refine_%05d" % step)
# coarse_checkpoint_path = COARSE_DIR + '/model.ckpt'
# saver_coarse.save(sess, coarse_checkpoint_path, global_step=step)
print("saving refine network checkpoint")
refine_checkpoint_path = REFINE_DIR + '/model.ckpt'
saver_refine.save(sess,
refine_checkpoint_path,
global_step=step)
else:
output_predict(logits_val, images_val,
"data/perdict/predict_%05d" % step)
print("saving coarse network checkpoint")
coarse_checkpoint_path = COARSE_DIR + '/model.ckpt'
saver_coarse.save(sess,
coarse_checkpoint_path,
global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()
def train():
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
dataset = DataSet(BATCH_SIZE)
images, depths, invalid_depths = dataset.csv_inputs(TRAIN_FILE)
keep_conv = tf.placeholder(tf.float32)
keep_hidden = tf.placeholder(tf.float32)
if REFINE_TRAIN:
print("refine train.")
coarse = model.inference(images, keep_conv, trainable=False)
logits = model.inference_refine(images, coarse, keep_conv,
keep_hidden)
else:
print("coarse train.")
logits = model.inference(images, keep_conv, keep_hidden)
loss = model.loss(logits, depths, invalid_depths)
train_op = op.train(loss, global_step, BATCH_SIZE)
init_op = tf.global_variables_initializer(
) #tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=LOG_DEVICE_PLACEMENT))
sess.run(init_op)
# parameters
coarse_params = {}
refine_params = {}
if REFINE_TRAIN:
for variable in tf.global_variables(): #tf.all_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
print("parameter: %s" % (variable_name))
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
else:
for variable in tf.trainable_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
# define saver
print(coarse_params)
saver_coarse = tf.train.Saver(coarse_params)
if REFINE_TRAIN:
saver_refine = tf.train.Saver(refine_params)
# fine tune
if FINE_TUNE:
coarse_ckpt = tf.train.get_checkpoint_state(COARSE_DIR)
if coarse_ckpt and coarse_ckpt.model_checkpoint_path:
print("Pretrained coarse Model Loading.")
saver_coarse.restore(sess, coarse_ckpt.model_checkpoint_path)
print("Pretrained coarse Model Restored.")
else:
print("No Pretrained coarse Model.")
if REFINE_TRAIN:
refine_ckpt = tf.train.get_checkpoint_state(REFINE_DIR)
if refine_ckpt and refine_ckpt.model_checkpoint_path:
print("Pretrained refine Model Loading.")
saver_refine.restore(sess,
refine_ckpt.model_checkpoint_path)
print("Pretrained refine Model Restored.")
else:
print("No Pretrained refine Model.")
# train
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
for step in range(MAX_STEPS):
index = 0
for i in range(1000):
_, loss_value, logits_val, images_val = sess.run(
[train_op, loss, logits, images],
feed_dict={
keep_conv: 0.8,
keep_hidden: 0.5
})
if index % 10 == 0:
print("%s: %d[epoch]: %d[iteration]: train loss %f" %
(datetime.now(), step, index, loss_value))
assert not np.isnan(
loss_value), 'Model diverged with loss = NaN'
if index % 500 == 0:
if REFINE_TRAIN:
output_predict(
logits_val, images_val,
"data/predict_refine_%05d_%05d" % (step, i))
else:
output_predict(logits_val, images_val,
"data/predict_%05d_%05d" % (step, i))
index += 1
if step % 5 == 0 or (step * 1) == MAX_STEPS:
if REFINE_TRAIN:
refine_checkpoint_path = REFINE_DIR + '/model.ckpt'
saver_refine.save(sess,
refine_checkpoint_path,
global_step=step)
else:
coarse_checkpoint_path = COARSE_DIR + '/model.ckpt'
saver_coarse.save(sess,
coarse_checkpoint_path,
global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()
def test():
with tf.Graph().as_default():
dataset = DataSet(BATCH_SIZE)
if (TEST_SINGLE):
images, depths, invalid_depths, filenames, depth_filenames = dataset.csv_inputs_test_single(
SINGLE_TEST_FILE)
else:
images, depths, invalid_depths, filenames, depth_filenames = dataset.csv_inputs_test(
TEST_FILE)
keep_conv = tf.placeholder(tf.float32)
keep_hidden = tf.placeholder(tf.float32)
if REFINE_TEST:
print("refine test.")
coarse = model.inference(images, keep_conv, trainable=False)
logits = model.inference_refine(images,
coarse,
keep_conv,
keep_hidden,
trainable=False)
else:
print("coarse test.")
logits = model.inference(images, keep_conv, trainable=False)
loss = model.loss(logits, depths, invalid_depths)
init_op = tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=LOG_DEVICE_PLACEMENT))
sess.run(init_op)
# parameters
coarse_params = {}
refine_params = {}
if REFINE_TEST:
for variable in tf.all_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
print("parameter: %s" % (variable_name))
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
else:
for variable in tf.all_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
# define saver
print coarse_params
saver_coarse = tf.train.Saver(coarse_params)
if REFINE_TEST:
saver_refine = tf.train.Saver(refine_params)
# load parameters
coarse_ckpt = tf.train.get_checkpoint_state(COARSE_DIR)
if coarse_ckpt and coarse_ckpt.model_checkpoint_path:
print("Pretrained coarse Model Loading.")
saver_coarse.restore(sess, coarse_ckpt.model_checkpoint_path)
print("Pretrained coarse Model Restored.")
else:
print("No Pretrained coarse Model.")
if REFINE_TEST:
refine_ckpt = tf.train.get_checkpoint_state(REFINE_DIR)
if refine_ckpt and refine_ckpt.model_checkpoint_path:
print("Pretrained refine Model Loading.")
saver_refine.restore(sess, refine_ckpt.model_checkpoint_path)
print("Pretrained refine Model Restored.")
else:
print("No Pretrained refine Model.")
# test
#Start populating the filename queue
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
if (TEST_SINGLE):
MAX_TEST_STEPS = test_single_number
for step in xrange(MAX_TEST_STEPS):
ture_depths, logits_val, images_val, loss_val, filenames_val, depth_filenames_val = sess.run(
[depths, logits, images, loss, filenames, depth_filenames],
feed_dict={
keep_conv: 0.8,
keep_hidden: 0.5
})
if REFINE_TEST:
if (TEST_SINGLE):
output_predict_test_single(
ture_depths, logits_val, images_val, filenames_val,
depth_filenames_val, "output/test/single_test_refine",
step * BATCH_SIZE)
else:
output_predict_test(ture_depths, logits_val, images_val,
filenames_val, depth_filenames_val,
"output/test/test_refine",
step * BATCH_SIZE)
else:
if (TEST_SINGLE):
output_predict_test_single(ture_depths, logits_val,
images_val, filenames_val,
depth_filenames_val,
"output/test/single_test",
step * BATCH_SIZE)
else:
output_predict_test(ture_depths, logits_val, images_val,
filenames_val, depth_filenames_val,
"output/test/test", step * BATCH_SIZE)
print("%s: %d[step]: test loss %f" %
(datetime.now(), step, loss_val))
assert not np.isnan(loss_val), 'Model diverged with loss = NaN'
coord.request_stop()
coord.join(threads)
sess.close()
def train():
'''
Train
'''
with tf.Graph().as_default():
# globalなstep数
global_step = tf.Variable(0, trainable=False)
# NYU Dataset V2 original size(480 x 640 x 3) -> crop -> (460 x 620 x 3)
image_input = ImageInput('./data/nyu_depth_v2_labeled.mat')
print("the number of train data: %d" % (len(image_input.images)))
images = tf.placeholder(tf.float32, [None, FLAGS.crop_size_height, FLAGS.crop_size_width, FLAGS.image_depth])
depths = tf.placeholder(tf.float32, [None, 1, 55, 74])
invalid_depths = tf.placeholder(tf.float32, [None, 1, 55, 74])
keep_conv = tf.placeholder(tf.float32)
keep_hidden = tf.placeholder(tf.float32)
# graphのoutput
if FLAGS.refine_train:
print("refine train.")
logits = model.inference_refine(images, keep_conv, keep_hidden)
else:
print("coarse train.")
logits = model.inference(images, keep_conv, keep_hidden)
# loss graphのoutputとlabelを利用
loss = model.loss(logits, depths, invalid_depths)
# 学習オペレーション
train_op = op.train(loss, global_step)
# サマリー
summary_op = tf.merge_all_summaries()
# 初期化オペレーション
init_op = tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(log_device_placement=LOG_DEVICE_PLACEMENT))
# saver
#saver = tf.train.Saver(tf.all_variables())
sess.run(init_op)
# coarseとrefineを分けて保存
coarse_params = {}
refine_params = {}
if FLAGS.refine_train:
for variable in tf.all_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count("/") != 1:
print("ignore.")
continue
scope, name = variable_name.split("/")
target, _ = name.split(":")
if variable_name.find('coarse') >= 0:
print("coarse parameter: %s" % (variable_name))
coarse_params[variable_name] = variable
if variable_name.find('fine') >= 0:
print("refine parameter: %s" % (variable_name))
refine_params[variable_name] = variable
else:
for variable in tf.trainable_variables():
variable_name = variable.name
print("parameter: %s" %(variable_name))
if variable_name.find("/") < 0 or variable_name.count("/") != 1:
print("ignore.")
continue
scope, name = variable_name.split("/")
target, _ = name.split(":")
if variable_name.find('coarse') >= 0:
print("coarse parameter: %s" %(variable_name))
coarse_params[variable_name] = variable
if variable_name.find('fine') >= 0:
print("refine parameter: %s" %(variable_name))
refine_params[variable_name] = variable
# define saver
saver_coarse = tf.train.Saver(coarse_params)
saver_refine = tf.train.Saver(refine_params)
# fine tune
if FLAGS.fine_tune:
# load coarse paramteters
coarse_ckpt = tf.train.get_checkpoint_state(COARSE_DIR)
if coarse_ckpt and coarse_ckpt.model_checkpoint_path:
print("Pretrained coarse Model Loading.")
saver_coarse.restore(sess, coarse_ckpt.model_checkpoint_path)
print("Pretrained coarse Model Restored.")
else:
print("No Pretrained coarse Model.")
# load refine parameters
refine_ckpt = tf.train.get_checkpoint_state(REFINE_DIR)
if refine_ckpt and refine_ckpt.model_checkpoint_path:
print("Pretrained refine Model Loading.")
saver_refine.restore(sess, refine_ckpt.model_checkpoint_path)
print("Pretrained refine Model Restored.")
else:
print("No Pretrained refine Model.")
# TODO train coarse or refine (change trainable)
#if not FLAGS.coarse_train:
# for val in coarse_params:
# print val
#if not FLAGS.refine_train:
# for val in coarse_params:
# print val
# train refine
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# debug
# サマリーのライターを設定
#summary_writer = tf.train.SummaryWriter(TRAIN_DIR, graph_def=sess.graph_def)
#batches = image_input.get_batches(FLAGS.batch_size)a
#d = np.asarray(batches[0][0])
#print d.shape
#a = np.asarray(batches[0][1])
#print a.shape
#logits_val, logits_fine_val, loss_value = sess.run([logits, logits_fine, loss], feed_dict={images: batches[0][0], depths: batches[0][1], invalid_depths: batches[0][2], keep_conv: 1.0, keep_hidden: 1.0})
#print len(logits_val[0])
#print len(logits_fine_val[0])
#print loss_value
# max_stepまで繰り返し学習
for step in xrange(MAX_STEPS):
start_time = time.time()
previous_time = start_time
index = 0
batches = image_input.get_batches(FLAGS.batch_size)
vals = image_input.get_validation()
for batch in batches:
train = batch[0]
depth = batch[1]
ignore_depth = batch[2]
_, loss_value = sess.run([train_op, loss], feed_dict={images: train, depths: depth, invalid_depths: ignore_depth, keep_conv: 0.8, keep_hidden: 0.5})
if index % 10 == 0:
end_time = time.time()
duration = end_time - previous_time
num_examples_per_step = BATCH_SIZE * 10
examples_per_sec = num_examples_per_step / duration
print("%s: %d[epoch]: %d[iteration]: train loss %f: %d[examples/iteration]: %f[examples/sec]: %f[sec/iteration]" % (datetime.now(), step, index, loss_value, num_examples_per_step, examples_per_sec, duration))
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if index % 50 == 0:
output_vec, cost_value = sess.run([logits, loss], feed_dict={images: vals[0], depths: vals[1], invalid_depths: vals[2], keep_conv: 1.0, keep_hidden: 1.0})
print("%s: %d[epoch]: %d[iteration]: validation loss: %f" % (datetime.now(), step, index, cost_value))
if index % 100 == 0:
output_dir = "predicts_%05d_%08d" % (step, index)
print("predicts output: %s" % output_dir)
data_feed_inputs_nyu.output_predict(output_vec, output_dir)
previous_time = end_time
index += 1
# if index % 100 == 0:
# pass
# summary_str = sess.run(summary_op, feed_dict={images: train, labels: label, keep_conv: 0.8, keep_hidden: 0.5})
# # サマリーに書き込む
# summary_writer.add_summary(summary_str, step)
#
if step % 5 == 0 or (step * 1) == MAX_STEPS:
if FLAGS.refine_train:
refine_checkpoint_path = REFINE_DIR + '/model.ckpt'
saver_refine.save(sess, refine_checkpoint_path, global_step=step)
else:
coarse_checkpoint_path = COARSE_DIR + '/model.ckpt'
saver_coarse.save(sess, coarse_checkpoint_path, global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()
def train(REFINE_TRAIN):
BATCH_SIZE = 8
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
dataset = DataSet(BATCH_SIZE)
keep_conv = tf.placeholder(tf.float32)
images, depths, invalid_depths, features = dataset.csv_inputs(
TRAIN_FILE)
if REFINE_TRAIN:
print("refine train.")
coarse = model.inference(images, trainable=False)
logits = model.inference_refine(images, coarse,
keep_conv) #？？？这个 参数是什么
else:
print("coarse train.")
logits = model.inference(images)
loss = model.loss(logits, depths, invalid_depths)
train_op = op.train(loss, global_step, BATCH_SIZE)
init_op = tf.global_variables_initializer() #改了
# Session
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=LOG_DEVICE_PLACEMENT)) # 不打印设备分配日志
sess.run(init_op)
# parametersi
coarse_params = {} # 定义一个新的dict
refine_params = {}
if REFINE_TRAIN:
for variable in tf.all_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
print("parameter: %s" % (variable_name))
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
else:
for variable in tf.trainable_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
# define saver
print(coarse_params)
saver_coarse = tf.train.Saver(coarse_params)
if REFINE_TRAIN:
saver_refine = tf.train.Saver(refine_params)
# fine tune 微调。。。
if FINE_TUNE:
coarse_ckpt = tf.train.get_checkpoint_state(COARSE_DIR)
if coarse_ckpt and coarse_ckpt.model_checkpoint_path:
print(coarse_ckpt.model_checkpoint_path)
saver_coarse.restore(sess, coarse_ckpt.model_checkpoint_path)
else:
print("No Pretrained coarse Model.")
if REFINE_TRAIN:
refine_ckpt = tf.train.get_checkpoint_state(REFINE_DIR)
if refine_ckpt and refine_ckpt.model_checkpoint_path:
saver_refine.restore(sess,
refine_ckpt.model_checkpoint_path)
else:
print("No Pretrained refine Model.")
# train
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
lossli = []
lossli1 = []
for step in range(MAX_STEPS):
index = 0
lossli = []
print('-------------------------------')
for i in range(3000):
_, loss_value, logits_val, images_val = sess.run(
[train_op, loss, logits, images],
feed_dict={keep_conv: 0.8})
if i % 100 == 0:
print('[Epoch]:', step, '[iteration]:', i,
'[Train losses]:', loss_value)
lossli.append(loss_value)
index += 1
lossli1.append(np.mean(lossli))
if step % 5 == 0 or (step * 1) == MAX_STEPS:
if REFINE_TRAIN:
refine_checkpoint_path = REFINE_DIR + '/model.ckpt'
saver_refine.save(sess,
refine_checkpoint_path,
global_step=step)
else:
coarse_checkpoint_path = COARSE_DIR + '/model.ckpt'
saver_coarse.save(sess,
coarse_checkpoint_path,
global_step=step)
plt.figure()
plt.plot(lossli1)
plt.savefig("train_loss.jpg")
plt.xlabel("Epoch")
plt.ylabel("Train_loss")
plt.title("Train_Loss for Each Epoch")
coord.request_stop() #请求所有线程停止
coord.join(threads) #等待所有的线程完成
sess.close()
def test():
BATCH_SIZE = 1
with tf.Graph().as_default():
dataset = DataSet(BATCH_SIZE)
keep_conv = tf.placeholder(tf.float32)
images, depths, invalid_depths, features = dataset.csv_inputs(
TEST_FILE)
coarse = model.inference(images, trainable=False)
logits = model.inference_refine(images,
coarse,
keep_conv,
trainable=False)
loss1 = model.loss(coarse, depths, invalid_depths)
loss2 = model.loss(logits, depths, invalid_depths)
init_op = tf.global_variables_initializer() #改了
# Session
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=LOG_DEVICE_PLACEMENT)) # 不打印设备分配日志
sess.run(init_op)
coarse_params = {} # 定义一个新的dict
refine_params = {}
for variable in tf.all_variables():
variable_name = variable.name
#print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count("/") != 1:
continue
if variable_name.find('coarse') >= 0:
coarse_params[variable_name] = variable
#print("parameter: %s" %(variable_name))
if variable_name.find('fine') >= 0:
refine_params[variable_name] = variable
saver_coarse = tf.train.Saver(coarse_params)
saver_refine = tf.train.Saver(refine_params)
# fine tune 微调。。。
if FINE_TUNE:
coarse_ckpt = tf.train.get_checkpoint_state(COARSE_DIR)
if coarse_ckpt and coarse_ckpt.model_checkpoint_path:
#print(coarse_ckpt.model_checkpoint_path)
saver_coarse.restore(sess, coarse_ckpt.model_checkpoint_path)
refine_ckpt = tf.train.get_checkpoint_state(REFINE_DIR)
if refine_ckpt and refine_ckpt.model_checkpoint_path:
#print(refine_ckpt.model_checkpoint_path)
saver_refine.restore(sess, refine_ckpt.model_checkpoint_path)
# test
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
index = 0
ls1 = []
ls2 = []
print('\n', '---------Examples---------:')
for step in range(NumOfTest):
#print('-----------------------------------------')
loss_value1, loss_value2, logits_val, coarse_val, images_val, features_ = sess.run(
[loss1, loss2, logits, coarse, images, features],
feed_dict={keep_conv: 1})
ls1.append(loss_value1)
ls2.append(loss_value2)
if step % 1 == 0:
index = index + 1
print(features_, 'Coarse losses:', loss_value1,
'Refine losses:', loss_value2, '\n')
output_save(coarse_val, logits_val, images_val, index,
"data/test")
ls1m = np.mean(ls1)
ls2m = np.mean(ls2)
print('---------Testing Results--------:')
print('Coasre image mean losses:', ls1m)
print('Refine image mean losses:', ls2m)
coord.request_stop() #请求所有线程停止
coord.join(threads) #等待所有的线程完成
sess.close()
def train():
'''
Train
'''
with tf.Graph().as_default():
# globalなstep数
global_step = tf.Variable(0, trainable=False)
# NYU Dataset V2 original size(480 x 640 x 3) -> crop -> (460 x 620 x 3)
image_input = ImageInput('./data/nyu_depth_v2_labeled.mat')
print("the number of train data: %d" % (len(image_input.images)))
images = tf.placeholder(tf.float32, [
None, FLAGS.crop_size_height, FLAGS.crop_size_width,
FLAGS.image_depth
])
depths = tf.placeholder(tf.float32, [None, 1, 55, 74])
invalid_depths = tf.placeholder(tf.float32, [None, 1, 55, 74])
keep_conv = tf.placeholder(tf.float32)
keep_hidden = tf.placeholder(tf.float32)
# graphのoutput
if FLAGS.refine_train:
print("refine train.")
logits = model.inference_refine(images, keep_conv, keep_hidden)
else:
print("coarse train.")
logits = model.inference(images, keep_conv, keep_hidden)
# loss graphのoutputとlabelを利用
loss = model.loss(logits, depths, invalid_depths)
# 学習オペレーション
train_op = op.train(loss, global_step)
# サマリー
summary_op = tf.merge_all_summaries()
# 初期化オペレーション
init_op = tf.initialize_all_variables()
# Session
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=LOG_DEVICE_PLACEMENT))
# saver
#saver = tf.train.Saver(tf.all_variables())
sess.run(init_op)
# coarseとrefineを分けて保存
coarse_params = {}
refine_params = {}
if FLAGS.refine_train:
for variable in tf.all_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
print("ignore.")
continue
scope, name = variable_name.split("/")
target, _ = name.split(":")
if variable_name.find('coarse') >= 0:
print("coarse parameter: %s" % (variable_name))
coarse_params[variable_name] = variable
if variable_name.find('fine') >= 0:
print("refine parameter: %s" % (variable_name))
refine_params[variable_name] = variable
else:
for variable in tf.trainable_variables():
variable_name = variable.name
print("parameter: %s" % (variable_name))
if variable_name.find("/") < 0 or variable_name.count(
"/") != 1:
print("ignore.")
continue
scope, name = variable_name.split("/")
target, _ = name.split(":")
if variable_name.find('coarse') >= 0:
print("coarse parameter: %s" % (variable_name))
coarse_params[variable_name] = variable
if variable_name.find('fine') >= 0:
print("refine parameter: %s" % (variable_name))
refine_params[variable_name] = variable
# define saver
saver_coarse = tf.train.Saver(coarse_params)
saver_refine = tf.train.Saver(refine_params)
# fine tune
if FLAGS.fine_tune:
# load coarse paramteters
coarse_ckpt = tf.train.get_checkpoint_state(COARSE_DIR)
if coarse_ckpt and coarse_ckpt.model_checkpoint_path:
print("Pretrained coarse Model Loading.")
saver_coarse.restore(sess, coarse_ckpt.model_checkpoint_path)
print("Pretrained coarse Model Restored.")
else:
print("No Pretrained coarse Model.")
# load refine parameters
refine_ckpt = tf.train.get_checkpoint_state(REFINE_DIR)
if refine_ckpt and refine_ckpt.model_checkpoint_path:
print("Pretrained refine Model Loading.")
saver_refine.restore(sess, refine_ckpt.model_checkpoint_path)
print("Pretrained refine Model Restored.")
else:
print("No Pretrained refine Model.")
# TODO train coarse or refine (change trainable)
#if not FLAGS.coarse_train:
# for val in coarse_params:
# print val
#if not FLAGS.refine_train:
# for val in coarse_params:
# print val
# train refine
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# debug
# サマリーのライターを設定
#summary_writer = tf.train.SummaryWriter(TRAIN_DIR, graph_def=sess.graph_def)
#batches = image_input.get_batches(FLAGS.batch_size)a
#d = np.asarray(batches[0][0])
#print d.shape
#a = np.asarray(batches[0][1])
#print a.shape
#logits_val, logits_fine_val, loss_value = sess.run([logits, logits_fine, loss], feed_dict={images: batches[0][0], depths: batches[0][1], invalid_depths: batches[0][2], keep_conv: 1.0, keep_hidden: 1.0})
#print len(logits_val[0])
#print len(logits_fine_val[0])
#print loss_value
# max_stepまで繰り返し学習
for step in xrange(MAX_STEPS):
start_time = time.time()
previous_time = start_time
index = 0
batches = image_input.get_batches(FLAGS.batch_size)
vals = image_input.get_validation()
for batch in batches:
train = batch[0]
depth = batch[1]
ignore_depth = batch[2]
_, loss_value = sess.run(
[train_op, loss],
feed_dict={
images: train,
depths: depth,
invalid_depths: ignore_depth,
keep_conv: 0.8,
keep_hidden: 0.5
})
if index % 10 == 0:
end_time = time.time()
duration = end_time - previous_time
num_examples_per_step = BATCH_SIZE * 10
examples_per_sec = num_examples_per_step / duration
print(
"%s: %d[epoch]: %d[iteration]: train loss %f: %d[examples/iteration]: %f[examples/sec]: %f[sec/iteration]"
% (datetime.now(), step, index, loss_value,
num_examples_per_step, examples_per_sec, duration))
assert not np.isnan(
loss_value), 'Model diverged with loss = NaN'
if index % 50 == 0:
output_vec, cost_value = sess.run(
[logits, loss],
feed_dict={
images: vals[0],
depths: vals[1],
invalid_depths: vals[2],
keep_conv: 1.0,
keep_hidden: 1.0
})
print("%s: %d[epoch]: %d[iteration]: validation loss: %f" %
(datetime.now(), step, index, cost_value))
if index % 100 == 0:
output_dir = "predicts_%05d_%08d" % (step, index)
print("predicts output: %s" % output_dir)
data_feed_inputs_nyu.output_predict(
output_vec, output_dir)
previous_time = end_time
index += 1
# if index % 100 == 0:
# pass
# summary_str = sess.run(summary_op, feed_dict={images: train, labels: label, keep_conv: 0.8, keep_hidden: 0.5})
# # サマリーに書き込む
# summary_writer.add_summary(summary_str, step)
#
if step % 5 == 0 or (step * 1) == MAX_STEPS:
if FLAGS.refine_train:
refine_checkpoint_path = REFINE_DIR + '/model.ckpt'
saver_refine.save(sess,
refine_checkpoint_path,
global_step=step)
else:
coarse_checkpoint_path = COARSE_DIR + '/model.ckpt'
saver_coarse.save(sess,
coarse_checkpoint_path,
global_step=step)
coord.request_stop()
coord.join(threads)
sess.close()